Fix CDMA decoding of multipart UTF-16 SMS messages.

Recent changes to support CMAS over CDMA introduced a bug causing
an exception to be thrown when decoding multipart UTF-16 encoded
messages. This change fixes the exception by correctly subtracting
the header size from the number of bytes to decode. It also adds
more robust error handling to try to decode the maximum length
possible instead of throwing an exception if the length is still
larger than the user data length after subtracting the header.

This also fixes a bug in the encoder, which was padding the
UTF-16 user data to 16-bit alignment, which is incorrect (should
be padded to an 8-bit boundary). The code happened to work because
we always generated a UDH that was an even number of bytes
(including length) so the padding was a no-op. The decoder works
correctly.

Bug: 6939151
Change-Id: I4000fa2f4703b39e5ed7e5bd8490828303ef8979